openpi Remote Inference: Where Observation Transforms Belong
For openpi remote inference, keep image preprocessing and state normalization on the sides expected by the served checkpoint, then verify action-chunk timing and units before robot use.
Physical AI Lab
TACTO: What Tactile Rendering Can and Cannot Validate
TACTO renders vision-based tactile observations while an external engine supplies contact dynamics. Choose tests that separate image handling from physical fidelity.
AnySkin: Checking Policy Reuse After a Tactile Skin Replacement
AnySkin makes a tactile surface replaceable, but policy reuse still needs a matched check. Separate unloaded readings, contact response and task transfer.
VoxPoser: Turning Language into 3D Planner Objectives
VoxPoser turns instructions into 3D planner objectives. Trace grounding, value maps, waypoints and control, and identify what the released simulator does not supply.
SayCan: Why a Useful Robot Skill May Still Be Infeasible
SayCan combines a skill’s relevance to an instruction with its feasibility in the current state. Learn what those scores mean and why execution still needs verification.
RT-H: Choosing Language Corrections or Teleoperated Corrections
RT-H separates language-motion choices from detailed robot actions. Learn when a language correction addresses the error and when a trajectory demonstration is needed.
Where2Act: Why an Actionable Pixel Is Not a Complete Action
A promising point on an object is only part of an action. Where2Act separates actionability, orientation proposals and scores for articulated interactions.
PerAct vs Act3D: Two Ways to Represent 3D Robot Actions
PerAct uses a voxel representation; Act3D refines spatial search. Compare their precision and coverage decisions without mixing benchmark conditions.
VRB: Learning Robot Affordances from Human Video
VRB extracts contact and hand-motion cues from human video. Learn what that supervision contributes and what the robot interface must still supply.
RoboPoint: Turning a 2D Action Point into a Robot Target
RoboPoint predicts a useful image location. Valid depth, camera calibration and a defined motion primitive are still needed to make it a robot target.